HMC Diagnostics
Script for checking MCMC/HMC diagnostics - this might be better as an Rmd to just have one output for supplemental publication
Plethodon jordani
View Traceplots
Check Domain Specific Expectations
Check N for Truncation
The augmentation to marginalize N out as a latent discrete requires setting an upper bound, K, to loop through. If K is too small the posterior will be truncated. Need to check for every N.
## null device
## 1Check detection
In simulations abundance esimates are unreliable when detection gets below 20%
Check Divergences and Pairwise Correlations
Check Energy and Treedepth
Summarize Samples Sizes and Mixing
Effective samples sizes using rstan::monitor following Hoffman and Gelman (2014) to be more reliable and accurate (as in Monnahan et al. 2017) - UPDATE - now follow Vehtari et al. 2019.
effective sample sizes (should be > 100) - Aki Vehtari, Andrew Gelman, Daniel Simpson, Bob Carpenter, and Paul-Christian Bürkner (2019). Rank-normalization, folding, and localization: An improved R-hat for assessing convergence of MCMC. arXiv preprint arXiv:1903.08008.
## [1] 1.003609## [1] 1675## [1] 2564Posterior Predictive Checks
Examine posterior prredictions of total counts across all 5 visits
RMSE of posterior predictive
## [1] 5.070722Posterior predictive check for each visit
RMSE of posterior predictive for observations per visit
## [1] 2.956443RMSE of posterior predictive
## [1] 1.322162Desmognathus wrighti
View Traceplots
Check Domain Specific Expectations
Check N for Truncation
The augmentation to marginalize N out as a latent discrete requires setting an upper bound, K, to loop through. If K is too small the posterior will be truncated. Need to check for every N.
## null device
## 1Check detection
In simulations abundance esimates are unreliable when detection gets below 20%
Check Divergences and Pairwise Correlations
Check Energy and Treedepth
Summarize Samples Sizes and Mixing
Effective samples sizes using rstan::monitor following Hoffman and Gelman (2014) to be more reliable and accurate (as in Monnahan et al. 2017) - UPDATE - now follow Vehtari et al. 2019.
effective sample sizes (should be > 100) - Aki Vehtari, Andrew Gelman, Daniel Simpson, Bob Carpenter, and Paul-Christian Bürkner (2019). Rank-normalization, folding, and localization: An improved R-hat for assessing convergence of MCMC. arXiv preprint arXiv:1903.08008.
## [1] 1.00399## [1] 2346## [1] 2987Posterior Predictive Checks
Examine posterior prredictions of total counts across all 5 visits
RMSE of posterior predictive
## [1] 1.315241Posterior predictive check for each visit
RMSE of posterior predictive for observations per visit
## [1] 0.7504749RMSE of posterior predictive
## [1] 0.3356226Eurycea wilderae
View Traceplots
Check Domain Specific Expectations
Check N for Truncation
The augmentation to marginalize N out as a latent discrete requires setting an upper bound, K, to loop through. If K is too small the posterior will be truncated. Need to check for every N.
## null device
## 1Check detection
In simulations abundance esimates are unreliable when detection gets below 20%
Check Divergences and Pairwise Correlations
Check Energy and Treedepth
Summarize Samples Sizes and Mixing
Effective samples sizes using rstan::monitor following Hoffman and Gelman (2014) to be more reliable and accurate (as in Monnahan et al. 2017) - UPDATE - now follow Vehtari et al. 2019.
effective sample sizes (should be > 100) - Aki Vehtari, Andrew Gelman, Daniel Simpson, Bob Carpenter, and Paul-Christian Bürkner (2019). Rank-normalization, folding, and localization: An improved R-hat for assessing convergence of MCMC. arXiv preprint arXiv:1903.08008.
## [1] 1.003496## [1] 2495## [1] 3415Posterior Predictive Checks
Examine posterior prredictions of total counts across all 5 visits
RMSE of posterior predictive
## [1] 1.404227Posterior predictive check for each visit
RMSE of posterior predictive for observations per visit
## [1] 0.8349771RMSE of posterior predictive
## [1] 0.3734131